THE EFFECT OF SHILL BIDDING UPON PRICES: EXPERIMENTAL EVIDENCE

Transcription

1 THE EFFECT OF SHILL BIDDING UPON PRICES: EXPERIMENTAL EVIDENCE GEORGIA KOSMOPOULOU a,* AND DAKSHINA G. DE SILVA b a Department of Economic, Univerity of Oklahoma, Norman, OK b Department of Economic, Texa Tech Univerity, Lubbock, TX Abtract Thi paper explore, through a erie of experiment, the effect of hill bidding upon revenue and price in auction. We tudy the practice of hill bidding in a common value framework. Our finding are conitent with the theoretical prediction that, if bidder are aware of the poibility of eller participation in an auction, hill bidding lower profit on average. Shill bidding can alleviate the problem of the winner' cure by lowering the price and it can, thu, provide benefit to the bidder. Finally, even though there were too many bidder that ubmitted bid in thee auction, the number of entrant wa not affected by the poibility of eller participation, which i alo conitent with the theory. JEL Claification: C91, D44. Key Word: Auction; Shill Bidding; Entry. *Correponding Author. Tel.: ; fax: addre: georgiak@ou.edu (G. Komopoulou). We thank Charle Nouair, Ron Hartad, Jamie Krue, George Delta, Wijeuriya P. Dayawana, the participant of the North American meeting of the Economic Science Aociation and the participant of the eminar erie at Tulane Univerity for helpful dicuion and comment. The financial aitance of the Univerity of Oklahoma Reearch Council i gratefully acknowledged.

2 1. Introduction There i an increaing concern over hill bidding in Internet auction. Shill bidding (or hilling) occur when the eller of an item poe a a bidder and ubmit bid in an auction in an effort to raie it price. Auction ite pend large amount of money to prevent thi activity. The peritence of hilling can affect the popularity of Internet auction a effective trading mechanim. Incident of hilling have alo been reported in traditional Englih auction for many year (ee Caady (1967), and Lucking-Reiley (2000)). In their tudy of how auction affected trade at the beginning of the 19 th century, Engelbrecht-Wiggan and Nonnenmacher (1999) reveal that thi practice wa widepread at that time. A change in New York legilature introduced in 1817 created diincentive for hiller. A a reult, the activity ubided contributing to the city rapid growth. Shill bidding ha important implication for the ucce of market and tudying it effect i of direct policy relevance. Rothkopf and Hartad (1995) for example attributed the rarene of Vickrey auction outide financial market partly to the fear of cheating eller. In thi paper, we report the finding of experiment that tudy the practice of hill bidding. We invetigate it effect on the bidding behavior and revenue of the eller in common value auction. We choe thi framework becaue a large number of item auctioned on-line are collectible (uch a antique tamp and port memorabilia) and econd hand good whoe value i uncertain. 1 Bidder etimate how much thee item are worth baed on their private information and the behavior of the other bidder. In uch auction, the eller can enter bid to milead the actual bidder and generate aggreive bidding pattern. In the proce, he run the rik of developing a bad reputation that could perit in future tranaction. Bidder often expre worrie about eller pat upect behavior. The rapid increae of tranaction over the Internet will create many more opportunitie for hill bidding in the future becaue eller can eaily diguie their identitie. Depite their large pending on thi, many bidder have complained that auction ite are not doing all they can to dicourage fraud. They attribute the reluctance to take action againt ome eller to the fact that 1 According to Bajari and Hortacu (2002) approximately 50% of the liting on ebay, the mot popular auction ite, can be claified a collectible. Se alo Lucking-Reiley (2000) and Bajari and Hortacu (2004). 1

3 higher price will ultimately generate higher commiion. Do price really increae when bidder anticipate the behavior of the eller? The recent theoretical work by Chakraborty and Komopoulou (2004) how that hill bidding doe not benefit the eller and i ome cae it doe not benefit the auctioneer either. Thi theory triggered our interet in the practical conequence of thi action. Our experiment reveal that hill bidding lower both price and profit contrary to the eller' intention. In that ene, it alleviate the problem of the winner' cure in common value auction. The information collected from the data allowed u to explore entry deciion and the extent of obervational learning. There are important practical difficultie in uing empirical data to invetigate the effect of hilling on bidding behavior and profit. Detection of hill bidding i difficult in Internet auction. Such information whenever it i available, it i typically enitive and i kept confidential. It i alo impoible to determine bidder knowledge of the occurrence of the practice. We created a computerized auction experiment that allowed u to trace the eller' participation pattern. We could control the amount of information that i available to bidder by carefully announcing the poibility of eller participation according to profit expectation. The paper i tructured a follow. Section 2 outline the modeling framework and ome related literature. Section 3 decribe the experiment, while ection 4 report and evaluate the reult. Section 5 ummarize our finding. 2. Theoretical framework We conider an open acending auction in which the highet bidder i awarded the item at the econd-highet bid price. Each bidder receive an independent common value ignal i taking value from a uniform ditribution. The value of the item i defined a the average of the n bidder' ignal: 1 V = n n i i= 1. In thi auction environment (i) there i a reerve price, (ii) entry take place imultaneouly and (iii) bidder cannot reenter once they have exited. Bidder bid the expected value of the item conditional on winning and conditional on the information revealed by other bid a the auction 2

4 progree (ee among other Milgrom and Weber (1982)). Thi modeling framework ha been previouly ued in many theoretical and experimental paper. 2 For implicity, we aume that the eller ha no value for the item and no information that could be ueful to the bidder. He ha, however, the ability to participate in ome auction and ubmit hill bid if he find it beneficial. The bidder do not know whether the eller i hill bidding, but they have a common belief in the probability that he doe. Chakraborty and Komopoulou (2004) recently analyzed the poibility of hill bidding in common value auction. They howed that the bidder take into account the potential for eller participation at the auction and revie their bid accordingly. Their work make the following tetable prediction for our framework of analyi: (i) If bidder are aware of the poibility of eller participation in an auction, hill bidding make the eller wore off. Seller would prefer it if there were a well-etablihed, trict enforcement mechanim that make it impoible for them to participate. (ii) A mixed participation trategy on behalf of the eller hould reduce price in an auction Experimental deign 4 Subject participated in 20 eion that lated for one-and-a-half hour and conited of a erie of 18 auction each. 5 They were recruited from a wide cro-ection of undergraduate tudent at the Univerity of Oklahoma and each participated in one eion. In each auction, there were at maximum five potential player. Each player wa either a true bidder or a eller. 2 Experiment include Avery and Kagel (1997), Holt and Sherman (2000), and Goeree and Offerman (2002). Theoretical work include Alber and Hartad (1991), Bikhchandani and Riley (1991), Klemperer (1998), and Bulow, Huang, and Klemperer (1999). 3 Chakraborty and Komopoulou (2004) aume, in general, that the expected value of the item increae in the number of bidder with high etimate of the value. In Theorem 6, they how that the price can increae in a mixed trategy hill bid equilibrium if ex ante (1) at mot one bidder i expected to have a high ignal at the auction and/or (2) the etimate of the common value increae at an increaing rate with the participation of additional bidder. In a common value auction, you don't expect ex ante that only one bidder will value the item highly and everyone ele will have low etimate unle there i a rare negative correlation in the etimate of the value. When ignal are uniformly ditributed, a in thi cae, a mixed participation trategy hould reduce the price and the potential for hill bidding hould make the eller wore off. 4 For more information on the deign, pleae ee the intruction and the naphot from an experiment preented in Figure 1-12B in Appendix B. 5 We performed 22 auction, 4 of which were trial run to familiarize the ubject with the auction environment. 3

5 The aignment wa decided at the beginning of the eion from a random draw. Our intention wa to evaluate how bidder learn and how they adjut their bidding trategie to the eller' behavior. The experiment were completely computerized and in each eion there were two treatment: in one the eller could not participate and in the other he could. 6 Subject received intruction for the econd treatment only after the et of auction of the firt treatment were completed. The order of treatment wa changed in ome eion to check for robutne of our reult. (i) The auction format. In each auction, a ingle unit of a commodity wa old to the highet bidder at the econd-highet bid price. We followed the "acending clock deign" in which there i a digital clock on the creen. The clock tarted at a particular bid (reflecting the reerve price) and moved upward every 5 econd. Each bidder wa able to oberve the clock and the proce wa interactive. 7 (ii) The ditribution of value. The value of all item were determined the ame way. Prior to the auction, the bidder received ignal revealing partial information about the value of the object. The average of the oberved ignal determined the value. Each ignal wa an integer drawn from a uniform ditribution between 0 and 20. The ignal and the ditribution of ignal were the only information available ex ante to bidder that could help them make deciion about entry. The eller did not receive any information ex ante that could be relevant to the bidder. The eller' value wa zero. (iii) The intruction. Subject were given a et of intruction deigned to help them undertand the nature of each object for ale and how to calculate the common value (ee appendix B). The eller wa given the opportunity to bid only in ome auction and hi potential 6 In thoe auction, the eller could oberve the outcome of the bidding proce and learn a much a every bidder would. 7 The acending clock deign i ued widely in the literature (ee among other Kagel, Hartad and Levin (1987) and the Kagel (1995)). We did not ue the format that pecifie an ending time for thee auction for two reaon. Firt and foremot, in a common value auction the fixed ending time rule poed an incentive problem: bidder hould wait until the lat minute to avoid revealing their private information (ee Lucking-Reiley (2000)). According to Bajari and Hortacu (2004): If all bid arrive at the lat minute, a bidder will not be able to update hi belief about the common value V uing the bid of other, hence hi bidding deciion would be equivalent to that of a bidder in a ealed-bid econd price auction. It turn out, that theoretically the fixed ending time rule would render hill bidding ineffective in equilibrium ince the eller would run the rik of becoming the highet bidder of the item without having the benefit of raiing the expected price conditional on ale. The econd le important and more practical reaon (which i relevant in traditional Englih auction) i that the auction ended in a matter of minute. 4

6 for participation, wa announced to all bidder. The eller wa able to paively oberve the outcome of the experiment in all auction in which he had no ability to participate. The bidder were given an initial budget of $15 to participate in the auction 8. We had 100 participant overall in the 20 eion. Subject were given enough time to read the intruction that were ubequently read aloud to them. The intruction included example that illutrated how the auction worked, and how the ubject could determine their profit or loe. (iv) The bidding. At the beginning of each auction, a reerve price (a minimum acceptable bid) wa poted on the creen. After each bidder received hi ignal he had to decide whether or not to participate. When all bidder made their deciion the auction would tart. The digital clock would appear on each creen along with information about the remaining cah balance. By clicking on a button next to the clock, marked "Bid Here," a bidder would be able to top hi clock and determine hi dropout price. When a ole clock wa left active, the remaining bidder obtained the object at the price hown by the clock the moment the lat-but-one bidder withdrew from the auction. Subject were paid in cah at the end of the eion (See Figure 2-5B in appendix B). (v) The information feedback. The ite alo diplayed information about the item that wa auctioned off. At any point in time, when an auction wa underway, the bidding hitory (coniting of the dropout price of all the bidder) wa poted on the ite. We concealed the identity of the bidder from each other to avoid direct identification of the eller' bid at any given auction. After each round, the true value, the winning bid, profit (or lo) and updated cah balance were reported to all bidder on their creen. The number of active bidder wa not announced at the beginning, but it could be inferred from the number of dropout price at the concluion of each auction. Thi i becaue, in many auction (inclu ding all Internet auction), the actual number of participant i not known ex ante [ee alo the dicuion in McAfee and Vincent (1992)]. The deign of the auction experiment ha the characteritic of the acending clock deign, an auction environment analyzed among other by Chakraborty and Komopoulou (2004). Bidder 8 To cover for 'bankruptcie' and 'no-hower,' we had extra bidder (up to two) in each experimental eion. The active bidder in each eion were elected on a 'firt-come -firt-erve' bai. We did not oberve any bankruptcie. Extra bidder were paid a $5 how-up fee. 5

7 could oberve the drop out price of their competitor and learn from their bidding pattern. A bidder bid, they revealed private information about the object to be old. Therefore, the remaining bidder could update their information about the object and bid accordingly. In each eion, the poibility of eller participation wa carefully announced to the bidder. It wa made clear that thi wa the eller' choice and not a certainty. The payoff function and the feedback information on dropout price could help them formulate and update their belief about the eller' participation trategy. 9 Garvin and Kagel (1994) point out that bidder learn to adjut their trategie through their own experience and obervational learning. Cooper and Kagel (2003) emphaize the importance of getting feedback regarding the outcome of earlier auction. They conclude that thi information will help bidder adjut their judgmental failure. The data obtained from thi experiment allowed u to invetigate the effect of hill bidding on price and payoff and bidder' learning within a eion. We could invetigate if the equential format of thee auction ha any effect on the bidding pattern and whether obervational learning play an important role in formulating trategie. 4. Experimental Reult In order to familiarize the ubject with the auction proce and let them gain ome bidding experience, we did two dry run each time and alo ued up the firt two paid auction a trial experiment. From each eion, we collected information for data analyi from 18 auction, 9 with and 9 without the poibility of eller participation. We collected 81 obervation from each eion. We have a total of 1616 obervation from 359 auction. 10 From thi et of data, we could identify the winner of each auction, the econd-highet bid, the number of bidder per 9 A an alternative to the approach we took, we thought of having the program imulate the behavior of the eller o that we could explicitly announce the probability of participation, in anticipation of optimal bidding behavior. However, ince the bidder' behavior i not alway optimal (according to the experimental evidence in Kagel and Levin (1986) participant behavior ignificantly differ from the Nah equilibrium behavior), uch a eller' fixed trategy would not be optimal either. A a reult, any concluion drawn from uch an analyi would be uele with any light deviation from the Nah equilibrium bidding trategie. We decided intead to let the bidder and the eller make their own deciion and evaluate the outcome. 10 In one eion, however, an error occurred when a bidder entered an auction other than the one under way at the time. We did not oberve a winner at thi auction. The bidder' cah balance wa not updated and, a a reult, we omitted the auction from our analyi. 6

8 auction, the value, the eller' profit and participation pattern, each bidder' identity, profit, ignal, dropout price, and cah balance. Baed on thee obervation, we traced the participation pattern and the repone of the bidder to change in the bidding environment. We analyzed the effect of hill bidding on price and the variance of price and examined if the entry deciion of bidder wa optimal or not in thee auction. We alo examined the robutne of the reult to change in the environment Entry Deciion The number of ubject who participated and ubmitted bid at thee auction wa larger than the number predicted by economic theory but on par with other experimental finding (Kagel and Levin (1991)). In a common value auction, a bidder hould enter if the expected value of the item conditional on winning at the reerve exceed the reerve price. If the value i the average of the bidder' ignal, the expected value of the item conditional on winning for a bidder whoe ignal i i can be expreed a: E[ V i i + ( n 1)* j ] = n i 1 l= 0 i l 1, for i j. In our cae, with four potential bidder, the optimal reerve price i 5.5. It wa calculated to maximize profit when the bidder chooe their optimal bidding trategie. 11 At thi reerve, any bidder with a ignal greater than or equal to 9 hould enter the auction. The following table preent participation tatitic at two reerve price ued to do a comparative tudy. Table-1: Baic Participation Statitic Variable Reerve 5.5 Reerve 6.25 All Auction Auction without Auction with All Auction Aucti on without Auction with Average Number of Bidder (0.772) Seller (0.717) Seller (0.827) (0.663) Seller (0.707) Seller (0.614) Probability of Bidder participation (0.359) (0.367) (0.337) 0.9 (0.3) Probability of Seller ubmitting a Bid See Appendix A for more detail. 7

9 Ten eion were performed at a reerve of 5.5. Baed on the data collected in thee eion, we oberved that the number of bidder who ubmitted bid wa greater than expected by Exceive entry i likely to make hill bidding more profitable in our analyi than theory would predict. Baed on the fact that the current reerve lead to high participation rate, we decided to perform ten additional experimental eion at a marginally higher reerve price of 6.25 (ee McAfee and Vincent (1992) on the iue of updating the reerve price). 13 At thi reerve, optimally, only bidder with a ignal greater than or equal to 10 hould participate in the auction. The reult of the comparative tudy are preented in Table 1 and 2. The announcement of eller participation did not have a ignificant impact on the number of participant and that i in agreement with the theory. The number of legitimate bidder only changed from to at the reerve of 5.5 and from to at the reerve of Theory predict that the eller' potential to enter hould not have an overall tatitically ignificant effect on the probability to participate. The entry deciion hould be baed on the comparion of the reerve price with the expected value of the item conditional on winning at the reerve. Since the eller doe not have any valuable information to hare with the bidder, the calculation of the expected value remain the ame irrepective of hi ability to participate. The probability of bidder participation at the reerve of 5.5 wa when the eller wa not allowed to participate and when he wa. A tet of thee proportion reveal that their difference i tatitically inignificant (with a z tatitic of 0.258). The difference in the probability of bidder participation between the two treatment at the reerve of 6.25 i tatitically inignificant a well 12 In an attempt to uncover their entry trategie, we calculated two other meaure of dicrepancy between oberved and expected entry number. The one meaure how the difference between the actual number of bidder and the number that would have entered if they calculated their return baed on an etimate of the common value conditional on their ignal; thi difference i The other meaure provide the difference between the actual number of bidder and thoe with a ignal greater than 5. Thi wa calculated to invetigate whether bidder decided to enter baed on a naïve direct comparion of their ignal with the reerve. Thi lat trategy eem marginally the cloet approximation to the oberved behavior. The dicrepancy, however, between thee two number i till large (the difference i 0.538), which make thi trategy a well an unlikely candidate for equilibrium behavior. 13 Some reearcher have peculated that the thrill of playing might be the dominant factor affecting the entry deciion of ome bidder. Since there i no hard evidence on what motivate bidder to enter in larger proportion than the optimal trategy would dictate, we were reluctant to raie the reerve too much in our comparative tatitic exercie to avoid introducing a reerve that would be too high. A ub-optimal high reerve combined with hill bidding could artificially how that hill bidding i harmful. A ub-optimal low reerve, on the other hand, can only overemphaize the benefit of hill bidding. We kept the reerve low and till howed that hill bidding i harmful. 8

10 (with a z tatitic of ). The evidence in Table 1 i corroborated by the probit analyi, the reult of which are preented in Table 2. Table-2: Probit Regreion Reult for the Probability of Bidding Independent Variable All Auction Reerve 5.5 Reerve 6.25 Contant (0.163) (0.166) (0.200) (0.207) (0.281) (0.284) Signal 0.159* (0.012) 0.159* (0.012) 0.156* (0.015) 0.157* (0.015) 0.163* (0.019) 0.163* (0.019) Potential Profit or Lo in Previou Auction (0.017) (0.023) (0.027) Order of Auction in the Sequence (0.014) (0.014) (0.016) (0.016) (0.029) (0.029) Seller' Potential Entry (0.151) (0.151) (0.187) (0.188) (0.304) (0.305) Seller Participation in the Previou Auction (0.139) 0.234* (0.141) (0.178) (0.184) 1.242* (0.304) 1.243* (0.305) Reerve 0.218* (0.100) 0.216* (0.100) Obervation χ * Denote 95% ignificance and ** Denote 90% ignificance In Table 2, we examine the probability to ubmit a bid a a function of bidder and auctionpecific independent variable. Thi model allow u not only to tet for difference in the probability to ubmit a bid acro the two treatment but alo to account for other oberved pattern of behavior in the equence of auction. The detailed decription of the variable ued in thi regreion i in Appendix A. The reult of the analyi reveal that the probability of ubmitting a bid i higher the higher the ignal the bidder received. The eller' potential to enter and the obervation of profit in the previou auction had no tatitically ignificant effect on the probability to ubmit a bid in a current auction. Thee reult are conitent with theory. The obervation of eller participation in an auction did not dicourage entry in the ubequent auction. In online auction, reputation mechanim etablihed have hown that negative feedback ha advere effect but i not a huge deterrent of activity ince the number of 9

11 auction participant experience continuou growth (ee Bajari and Hortacu (2004)). Contrary to what we expected, however, participation wa greater at the higher reerve price. In concluion, depite the fact that bidder did not make optimal entry deciion, the number of entrant wa invariant to the poibility of eller participation The Effect of Shill Bidding on Price and Profit In thi ection, we firt preent a graph and baic tatitic on relative bid and profit. Then we ue the Maximum Likelihood Etimation procedure to tatitically analyze our data. SELLER'S PROFITS AND DROPOUT PRICES Relative Price MEAN OF THE SELLER'S RELATIVE PROFIT MEAN OF THE SELLER'S RELATIVE DROPOUT PRICE Auction Number Figure 1A: Seller' relative profit and bidding behavior when the reerve wa The potential of participation wa announced only in the lat 9 auction. Figure 1A how the eller' relative profit and dropout price in the erie of auction performed at the reerve of In every eion, we ran 9 conecutive auction without the poibility of eller participation followed by 9 conecutive auction with the poibility of eller participation. 14 The profit and price are calculated here relative to the value of each item. For 14 A figure preenting the profit of the eller at the reerve of 5.5 would not be informative becaue we changed the order of announcement in ome eion to check the robutne of our reult to variation in 10

12 every auction in the equence, the etimate preented in Figure 1A are obtained by averaging out the correponding quantitie acro the 10 eion. The graph reveal that eller' relative profit are higher, on average, in the firt 9 auction of each eion when the bidder are the only one participating. In the lat 9 auction, eller were mixing their participation trategie; on average they ubmitted bid with a probability of 74.4%. There wa no eller who either participated all the time or did not participate at all. Profit were lower when participation wa poible. The larger drop in expected profit occurred right after the announcement of eller participation wa made. The pattern in price i imilar to the pattern in profit. Thi i in agreement with the theoretical reult reported by Chakraborty and Komopoulou (2004) which how that, in the preent etting, the auctioneer (who care about the price) i wore off with a hill bid equilibrium than with the no hill bid outcome. Conidering the data from all the auction, on average, eller ubmitted bid with a probability of 73.3%. Their behavior led to lower profit there a well. Table-3: Relative price and profit Variable All Auction Auction without Seller Average Relative Winning Price for Bidder (0.357) (0.405) Auction with Seller (0.300) Average Relative Profit for Bidder (0.357) Average Relative Dropout Price for Seller Average Relative Profit for Seller (0.369) (0.405) (0.385) (0.300) (0.368) (0.347) In fact, according to Table 3, the average relative profit of the eller went down from 97.5% to 88.9% of the value. Our tet revealed that the difference i tatitically ignificant (t tatitic i ) The average relative winning price for the bidder at the ame time dropped lightly from the format. In half the eion, in auction 1-9, the eller had to be paive and in auction he could be potentially active. In the other half he had to be paive in auction 1-5 and and could be active in the remaining auction. 11

13 102.6% of the value to 98.1%. The difference in price acro the two treatment howed to be tatitically ignificant a well (t tatitic i 2.223). Once more, the main drop wa oberved right after the firt announcement of the potential for eller participation wa made. The bidder' profit were, on average, 2.6% of the value in the auction without eller participation and 1.9% of the value in the auction when the potential of eller participation wa announced. 15 The baic tatitic appearing in Table 3 do not provide any control for a variety of factor that affect price and profit. For that reaon, we ued the MLE procedure to identify the effect of hill bidding on the expected value and variance of price and profit. The baic tructure of the regreion model i a follow: y i = XB + ZΓ ε i ~ N(0, σ i i 2 + ε i, ), where σ = XΦ + ZΨ + γ i. We ue two dependent variable in our analyi: price and profit. The independent variable include two et of control: auction pecific quantitative variable (X) and, auction pecific qualitative variable (Z). We aume that the tandard deviation of price and profit depend on the ame et of variable to allow for more flexibility. The decription of thoe variable i detailed in Appendix A. We ued a variable on the oberved number of participant in the lat auction to etimate the aggreivene of opponent baed on oberved pattern of participation. We alo included the order of auction in the equence and the profit (or lo) in a previou auction. We accounted for the eller potential entry (hill bidding) and hi participatio n in a previou auction. We included difference in the deign, reerve, and value. In the profit equation, we alo introduced the variable Seller Win the Auction. Thi variable identifie the 15 Our analyi conider the overall effect of hill bidding on price and profit without concentrating on comparion of individual bid to equilibrium bid. We compare entry deciion to equilibrium behavior only where uch an approach i feaible. Notice that, once a bidder enter, each bid ubmitted i conditioned on information that i obtained by inverting the bidding function and uncovering the other bidder' ignal. Thi proce require knowledge on our behalf of the bidder' belief about the probability of the eller' participation. Thi i information we do not have. Even if we knew the belief, a direct comparion of actual bid to optimal bid would be quite mileading, ince a potential error in the calculation made by one bidder would propagate a erie of mitake in calculation by the ret. A a reult, ome optimally behaving bidder could eem to behave ub-optimally. 12

14 cae in which the eller win the auction and forgoe the price; when it i included in the model, it allow u to iolate through the variable on potential entry announcement the effect of eller participation on the price the bidder pay. Table 4 preent the reult of thi analyi. In thi analyi, we recognized that, if the ignal play an important role in determining the bidding trategy, then the tandard deviation of price could be related to the tandard deviation of ignal. The tandard deviation could alo be a function of the expected number of bidder in an auction, the eller' potential to enter, hi participation in previou auction, the level of the reerve price, and the potential for profit. Table 4 preent a maximum likelihood etimation of price and profit a a function of the obervable factor decribed above. It ugget that the value of the item ha a poitive effect on the price. The number of bidder in the previou auction puhe the price up becaue it created expectation for pattern of greater participation. Thi i not conitent with the theory, but it i conitent with other experimental finding. In particular, Kagel and Levin (1986) find that, a the number of bidder increae, the price alo increae. The potential of eller participation affect bidder belief and make them bid and win at lower price. The obervation of eller hilling in previou auction ha a negative but marginally ignificant effect on price. Obervation of pat performance through variou reputation mechanim etablihed in web auction have hown that negative feedback can have advere effect. Thoe effect vary and may depend alo on the value of the object old (Ba and Pavlou, (2002)). The bidder bid higher and higher in the equence of item auctioned off in a eion. There have been many tudie documenting increaing and decreaing pattern of price in equential auction. The mot relevant tudy for the preent framework, however, i that by Milgrom and Weber (2000) who predict that, in a model with affiliated value (an extreme form of which i having common value), there will be an increaing pattern of price in the equence. The change in the reerve price doe not eem to have a ignificant effect on the average price but it ha an effect on the tandard deviation of price. The higher reerve lead to lower variability in price. The order of the treatment doe not affect the outcome. 13

15 Table-4: ML Etimation Reult Independent Variable Dependent Variable Price Profit Contant 3.895* (0.601) 6.809* (0.566) 3.833* (0.671) 3.960* (0.607) Value 0.295* (0.036) 0.323* (0.038) 0.282* (0.036) Number of Bidder in the Previou Auction 0.424* (0.127) 0.426* (0.138) 0.268** (0.144) 0.360* (0.123) Potential Profit or Lo in Previou Auction (0.041) (0.046) (0.044) (0.039) Order of Auction in the Sequence 0.129* (0.026) 0.122* (0.028) 0.106* (0.031) 0.109* (0.025) Seller' Potential Entry * (0.279) * (0.306) * (0.304) * (0.267) Seller Participation in the Previou Auction ** (0.283) (0.283) (0.367) (0.276) Deign (0.316) (0.367) (0.335) (0.293) Reerve (0.222) (0.281) (0.234) (0.204) Seller Win the Auction * (0.676) Parameter Etimate of σ v i Contant 1.784* (0.433) 2.178* (0.503) (0.592) 1.779* (0.519) Standard Deviation of the Signal (0.035) (0.041) (0.049) (0.036) Number of Bidder in the Previou Auction (0.097) (0.120) (0.110) (0.100) Potential Profit or Lo in Previou Auction (0.034) (0.037) (0.042) (0.033) Order of Auction in the Sequence (0.026) (0.030) (0.037) (0.028) Seller' Potential Entry (0.293) (0.363) (0.524) (0.309) Seller Participation in the Previou Auction (0.246) (0.265) (0.526) (0.241) Deign (0.260) (0.252) (0.396) 0.442** (0.246) Reerve (0.224) * (0.226) (0.286) (0.202) Seller Win the Auction (0.279) Obervation Wald χ 2 Log Likelihood * Denote 95% ignificance and ** Denote 90% ignificance. 14

16 Table 4 offer no clear indication of ome ignificant poitive effect of the tandard deviation of ignal on the tandard deviation of the price. The eller' potential entry deciion would be expected to lower the variability in the price ince it could lead to le aggreive bidding behavior. Thi effect i, however, tatitically inignificant. The eller' profit i higher with the value of the item to the bidder, the number of bidder oberved in the previou auction, and the order of auction in the equence. The announcement of the potential of eller participation ha a negative effect on profit not only becaue the eller run the rik of buying hi own item but alo becaue the bidder pay a lower price on average. In particular, when the variable Seller Win the Auction i introduced in the profit equation (in the fourth column of Table 4) to iolate the intance in which the eller i awarded the item, the variable Seller' Potential Entry remain ignificant. A Chakraborty and Komopoulou (2004) have hown, hill bidding make the eller wore off. Our experimental reult agree with their finding. Any out-of-auction mechanim that would force eller to abtain from participation could increae their profit. Rothkopf and Hartad (1995) preented a theoretical dynamic model of cheating in Vickrey auction howing alo ignific ant advere effect. In their cae, bidder are not fully rational and they punih cheating whenever it can be verified. A truted eller cheat when it pay and eventually detroy hi reputation. The increae in the reerve doe not have a tatitically ignificant effect on either the expected value or the tandard deviation of profit. In Table 5, we included variable to capture the actual participation pattern of eller. One variable control for intance in which the eller i not allowed to participate, and another intance in which he i actually participating (a fact that i not obervable by the ret of the bidder). The control group repreent the cae in which the eller had the opportunity but choe to abtain from participation. Once more, the regreion reult ugget that the eller i better off with an enforcement mechanim that reduce hi ability to participate. In particular, hi payoff i the lowet when he enter and bid in the auction. We alo teted the robutne of our reult to change in the order of announcement. The MLE reult indicate that it doe not make any tatitically ignificant difference in either the price or the eller' profit. The eller' revenue i lower when the potential of participation i announced. In all equation, the obervation of profit or lo in the previou auction doe not eem to affect either price or profit. 15

17 Table-5: ML Regreion Reult Independent Variable Dependent Variable Profit Contant 3.489* (0.817) 3.734* (0.726) Value 0.324* (0.038) 0.275* (0.035) Number of Bidder in the Previou Auction 0.309* (0.126) 0.381* (0.110) Potential Profit or Lo in Previou Auction (0.042) (0.038) Order of Auction in the Sequence 0.101* (0.030) 0.113* (0.024) Seller i not allowed to participate (0.352) (0.337) Seller i Participating in the Auction * (0.315) * (0.298) Deign (0.314) (0.285) Reerve ( (0.206) Seller Win the Auction * (0.721) Parameter Etimate of σ v i Contant (0.820) 1.238* (0.695) Standard Deviation of the Signal (0.046) (0.034) Number of Bidder in the Previou Auction (0.120) (0.111 Potential Profit or Lo in Previou Auction (0.040) (0.032) Order of Auction in the Sequence (0.035) (0.027) Seller i not Participating in the Auction (0.346) (0.312) Seller i Participating in the Auction (0.431) (0.251) Deign 0.753* (0.363) 0.412** (0.248) Reerve (0.263) (0.191) Seller Win the Auction (0.254) Obervation Wald χ 2 Log Likelihood * Denote 95% ignificance and ** Denote 90% ignificance. 16

18 5. Concluion Thi paper examine the effect of hill bidding in online auction on the eller' payoff and on price. Shill bidding make the eller wore off. The price at the auction decreae a bidder anticipate the behavior of the eller and adjut their bidding trategie. Both of thee finding are conitent with the theoretical reult of Chakraborty and Komopoulou (2004). Even though their entry deciion i invariant to the announcement of eller participation, there are too many bidder who are eager to ubmit bid at thee auction. Obervational learning play a role in determining bidding trategie. The obervation of conitently large pat participation affect their bidding. A Milgrom and Weber (2000) predicted, overall there i an increaing pattern of price in the equence. We conclude that the poibility of hill bidding alleviate the problem of the winner' cure in a common value auction and become beneficial for bidder but harmful for the eller. 17

19 Reference Alber, W. and Hartad R.M., 1991, Common value auction with independent information: a framing effect oberved in a market game, in (R. Selten, ed.) Game Equilibrium Model, vol. 2, Berlin: Springer-Verlag. Avery, Chritopher and John H. Kagel, 1997, Second-Price Auction with Aymmetric Payoff: An Experimental Invetigation, Journal of Economic and Management Strategy, 6(3): Ba, Sulin and Paul Pavlou, 2002, Evidence of the Effect of Trut Building Technology in Electronic Market: Price Premium and Buyer Behavior, MIS Quart. 26, 3, pp Bajari, Patrick and Ali Hortacu, 2002, Cyberpace Auction and Pricing Iue: A Review of Empirical Finding, WorkingPaper 02005, Stanford Univerity, Department of Economic., 2004, Economic Inight from Internet Auction, Journal of Economic Literature, v. 42, 2, pp Bikhchandani, S. and Riley, J.G., 1991, Equilibria in open common value auction, Journal of Economic Theory, vol. 53, pp Bulow, J.I., M. Huang, and P. D. Klemperer, 1999, Toehold and takeover, Journal of Political Economy, vol. 107, pp Caady, R., Jr., 1967, Auction and Auctioneering. Univerity of California Pre, Berkeley. Chakraborty, Indranil and Georgia Komopoulou, 2004, Auction with Shill Bidding, Economic Theory, 24, Cooper, David and John H. Kagel, 2003, Leon Learned: Generalizing Learning Acro Game, American Economic Review, 93, Engelbrecht-Wiggan, R. and Nonnenmacher, T., 1999, 'A Theoretical Bai for 19th Century Change to the Port of New York Imported Good Auction', Exploration in Economic Hitory, 36, pp Garvin, Suan and John H. Kagel, 1994, Learning in Common Value Auction: Some Initial Obervation, Journal of Economic Behavior and Organization, 25, Goeree, Jacob K. and Theo Offerman, 2002, Efficiency in Auction with Private and Common Value: An Experimental Study, American Economic Review, 92(3): Holt, Charle, A. and Roger Sherman, 2000, Rik averion and the winner' cure, working paper. Kagel John H, 1995, Auction: A Survey of Experimental Reearch in (J.H. Kagel and A.E. Roth ed.) Handbook of Experimental Economic, Princeton Univerity Pre, Princeton, New Jerey. 18

20 Kagel, John H.; Hartad, Ronald M.; Levin, Dan, 1987, Information Impact and Allocation Rule in Auction with Affiliated Private Value: A Laboratory Study, Econometrica, 55(6), Kagel, John H. and Dan Levin, 1986, The winner cure and Public information in Common Value auction, The American Economic Review, 76(5), , 1991, The winner cure and Public information in Common Value auction The American Economic Review, 81(1), Klemperer P.D., 1999, Auction with almot common value, European Economic Review, vol. 42, pp Lucking-Reiley David, 2000, Auction on the Internet: What Being Auctioned and How? Journal of Indutrial Economic, vol. 48, 3, pp McAfee, Preton R., and Vincent, Daniel, 1992, Updating the reerve price in common-value auction, The American Economic Review, 82, Milgrom, Paul R., and Robert J. Weber, 1982, A Theory of Auction and Competitive Bidding, Econometrica, 50, Milgrom, Paul R., and Robert J. Weber, 2000, A Theory of Auction and Competitive Bidding II, In Klemperer P. (ed.) The Economic Theory of Auction, Cheltenham, UK: Edward Elgar, vol. II pp Rothkopf, Michael H., and Ronald M. Hartad, 1995, Two Model of Bid-Taker Cheating in Vickrey Auction, Journal of Buine, 68,

21 20 Appendix A CALCULATION OF THE OPTIMAL RESERVE We need to determine the minimum entering ignal that allow the eller to maximize hi profit given the bidding trategie. The expected price at the auction conditional on thi cutoff ignal i: = + < + < + < = ) ( ) ( ) 21 ( ) ( ) 21 5( ] [ ] [ ] [ ] [ ] [ y P E y y P E y y P E y y P E p E where ,,, y y y y are order tatitic. Chooing to maximize thi expreion yield a value of = Since we have dicrete ignal = 9. Baed on thi we can calculate the optimal reerve to be = + = = i i r. Since the clock goe up in increment of 0.25, we et the reerve at 5.5.

22 DEFINITIONS OF THE VARIABLES Variable Price Seller Profit Value Number of Bidder in the Previou Auction Potential Profit or Lo in Previou Auction Order of Auction in the Sequence Seller' Potential Entry Seller Participation in the Previou Auction Deign Decription and Contruction of the Independent Variable The Price i the drop out price of the econd highet bidder. Seller profit i the price minu a commiion of 5% that goe to the auctioneer a payment for the ervice rendered. If a eller doe not participate at all or if he participate and he i not the highet bidder, then the eller will receive the price minu the commiion. If eller become the highet bidder in an auction then he will not receive any payment on thi item but he will till have to pay 5% a the commiion even if the item i not old to an actual bidder. Thi i the value of the item to the bidder. Thi value i contructed a the average of the bidder' ignal. Thi variable repreent the oberved number of participant in the lat auction. It allow u to control for the aggreivene of opponent baed on oberved pattern of participation. Thi variable capture the difference between the value and the price that make up the profit or lo accruing to the winner of the previou auction. The variable capture effect that relate to the equential nature of the auction in each eion. Thi i a dummy variable that take the value of 1 in auction in which the potential of participation i announced and zero when the eller i not allowed to participate. Thi i dummy variable, which take the value of 1 if there were 5 bidder in the previou auction and therefore, the bidder oberved with certainty the hiller bidding and zero otherwie Thi dummy variable control for the change in order of the treatment. For 5 of the experimental eion performed at the reerve of 5.5 and 10 performed at the reerve of 6.25 the rule were a follow: the eller had to be paive in the firt half of the auction that took place. For 5 eion performed at the reerve of 5.5, we changed the order of announcement to check if our reult are robut. The eller had to be paive in auction 1-5 and and he could be active in the remaining auction. For thoe five eion the deign dummy take the value of 1 and 0 in all other eion Reerve Thi i a dummy that take the value of 1 when the reerve i 6.25 and zero when it i

23 Seller Win the Auction The Seller i not Allowed to Participate The Seller i Participating Standard Deviation of the Signal Thi i a dummy variable which take the value 1 when eller win the auction. Thi variable i introduced only in the profit equation to control for the intance in which the eller i awarded the item. Further, it allow the variable on potential entry announcement to iolate the effect of eller participation on the price the bidder pay. Thi variable take the value of 1 in auction in which we announced that the eller could not participate. Thi variable i ued only when the eller profit i analyzed. Thi i a dummy variable that take the value of 1 if the eller participated in an auction. Note that, the ret of the bidder do not oberve when the eller actually participate. Thi variable wa ued only for a eller profit ML etimation. In each auction bidder received 4 ignal. The tandard deviation of thee four ignal i ued in the maximum likelihood variance equation. Thi variable will be controlling for ignal variability. 22

24 Appendix B (REFEREE APPENDIX) INSTRUCTIONS (The intruction were read to all participant.) I. INITIAL SET OF INSTRUCTIONS Intruction to the bidder: Welcome to the experiment! We will hold a erie of online auction following the ame rule each time. The intruction are imple. If you read them carefully, take into account the reaoning of the other player and decide enibly, you will make ome money. Your profit depend on your ucce. Participation i voluntary. You are by no mean obliged to participate in the experiment but if you do, you will get the chance to make ome money if you make the right deciion. For each point that you will obtain in the experiment you will receive a quarter. The Game. The game will be played by group of 4 people. If you decide to participate you will receive a tarting balance of 60 point (That' $15). We will auction off 22 item, one at a time. Each item will be auctioned off to the highet bidder. The rule are lightly different than the rule of the tandard auction that you ee online. Here are the difference: The Value. Every one of the 22 item ha a different value. The value of each item i determined a follow: Each peron will receive a ignal at the beginning of each auction. The ignal could be any integer between 0 and 20. All thee number are equally likely. A peron only know hi/her own ignal. The value of the item i the ame for all bidder; it i the average of the ignal received by the 4 bidder. For example, if you received a ignal of 0 and the ignal received by the ret of the bidder are 9, 6, and 17, the common value of the item to all bidder will be: = 8. Your ignal will be hown at the top border of your creen. You 4 will alway get information about your own ignal. You will not get to know the ignal of the other bidder before the end of the auction. When all member of a group receive their ignal, the object i auctioned off. The Rule of the Auction. When the auction begin a reerve price (a minimum acceptable bid) will be poted on your creen. After you receive your ignal you will have a minute to decide whether to participate. If you decide to participate you have to pre the button that ay "participation" and wait for the ret of the bidder to make a deciion. Thi i the only chance you will get to decide whether you are going to participate and bid in thi particular auction. (If 23

25 you don't find it profitable to participate in thi auction, baed on the information that you received on thi item, you can till participate in the next round of auction.) The value of the item depend on the ignal of all 4 people no matter whether they decided to participate or not. In the middle of the creen, you will ee a button that how a bid lowly counting upward like a clock. Every active participant ee the ame clock on hi or her creen. When you pre thi button on your creen that ay "Bid Here" your clock with top counting up and you will leave the auction. The bid at which you preed the button i called the "dropout price". The auction will continue with the participant that have not yet preed their own button. When only one peron remain in the auction, thi peron leave automatically and obtain the object at the price that i currently indicated i.e., the dropout price of the econd highet bidder that left the auction. For example, if there are two bidder remaining active in the auction and the one decide to dropout at the price of 7, the other bidder i awarded the item at 7. The peron that manage to obtain the object receive a certain amount of point on hi or her account; thi amount i determined by the common value of the item to all bidder minu the bid of the econd highet bidder that left the auction. In the previou example, if the (common) value of the item i 8 (i.e., the average of the ignal of all four bidder) and the dropout price of the econd highet bidder i 7 then the highet bidder will earn a profit of 8-7=1 that will be added to hi account. If the dropout price of the econd highet bidder wa 11, however, the peron that i awarded the item will loe 11-8 = 3 that will be ubtracted from hi account balance. Bankruptcy Policy. A we mentioned before, you will have a 60-point balance in you account available for bidding. Thi i the maximum amount of point you can ue bidding in thee auction. If at ome point of time you loe all point overbidding on a erie of item you will not get a chance to reenter and bid in ubequent round. Ueful Information on the Screen. A you decide on your trategy take into account the behavior of other participant. The dropout price of the other participant are reported on the creen in the column that ha the title bidding hitory. A oon a any player leave the auction hi dropout price become an entry in the bidding hitory. At the end of every round your remaining cah balance and the profit or lo will be diplayed on the creen. You will alo ee the winning bid and the value of the item to the bidder. To enter the experiment logon to Good luck! 24

26 Intruction to the eller: For the firt 12 round you will have the opportunity to oberve the outcome in each auction, the bid and value of the item auctioned off. After the end of the firt 12 auction in the eion, the rule will change and you will be given the opportunity to participate if you wih. You will receive further intruction at the end of round 12. Logon to /Georgia/page33.ap to oberve the bidding proce during the firt 12 auction. II. ADDITIONAL INSTRUCTIONS Intruction to Bidder: For the next et of auction the eller ha the opportunity to enter and bid if he find it beneficial. The eller doe not have any ignal that could convey additional information about the value of the item to you. Whether the eller decide to participate or not and hi/her identity will not be revealed during thee auction. Intruction to the Seller: Your value of the item i zero and it i independent of the value of other participant. Your initial cah balance will be 60 point. For each point that you will obtain in the experiment you will receive a dime. At thi point in the experiment you have the opportunity to enter and participate if you wih. Every time an item i auctioned off to one of the bidder, independent of your participation deciion, you will receive the price minu a commiion of 5% that goe to the auctioneer a payment for the ervice rendered. If you become the highet bidder in an auction you will not receive any payment on thi item but you will till have to pay 5% a the commiion even if the item i not old to an actual bidder. For example, if you are the lat active peron at the auction and the econd highet bidder dropped out at a price of 9 you will have to pay 5% of 9 which i If you do not participate at all or if you participate and you are not the highet bidder you will receive the price minu the commiion. For example, if the econd highet dropout price i 6 you will receive 6 and you will pay 5% of the price to the auctioneer. A a reult your net profit will be 5.7. The bankruptcy policy that applie to the bidder applie to you too. 25

27 SNAPSHOTS FROM AN EXPERIMENT Figure1B: Bidder' page. 26

28 Figure 2B: After a bidder enter the eion and auction number that i announced, he/he get to ee a ignal and make a deciion on participation. 27

29 Figure 3B: A bidder who decide to participate at the auction ee the digital clock on hi creen, information about hi cah balance, and the bidding hitory a the auction progree. 28

30 Figure 4B: A bidder drop out, their dropout price appear on the creen of active bidder. 29

31 Figure 5B: When the auction i concluded, the bidding hitory, the winning bid, the value of the item to the bidder, the profit or lo, and the remaining cah balance appear on the creen. 30

32 Figure 6B: A bidder who doe not participate at the auction get the ame information a every other bidder but the clock will no longer be available on hi creen. 31

33 Figure 7B: At the end of every auction, thoe who do not participate oberve the ame information on the creen a thoe who participate. 32

34 SELLER S OBSERVATION WINDOW Figure 8B: When the eller i not participating, he/he get to oberve the outcome of the auction, including the bid, from the following obervation window. All the information that i available to bidder i available to the eller a well (other than the bidder' ignal). 33